Automatic Testing for an NLP Solver by Generating Feasible or Infeasible Problems (LPs, QPs, OCPs)

Philippe Monmousseau

Ecole Polytechnique, France

Tuesday, June 10, 2014, 14:15 - 14:45

Room 01-012, Georges-Köhler-Allee 102, Freiburg 79110, Germany

There exist several nonlinear optimization problem solvers, and we use them in different applications. However, there is no generic way of ensuring oneself that a solver is able to deal with every problem. This presentation’s aim is to introduce several methods for testing those solvers by studying ways of generating problems of which we know beforehand if they are feasible or not. These methods were developed during a summer internship at the Control and Optimization Laboratory of the University of Freiburg, and tested on the Dynobud software (written by PhD student Greg Horn) which uses the IPOPT solver.